CN102607867B

CN102607867B - On-passage fault detection system based on GLRT (generalized likelihood ratio test) train suspension system and detection method of on-passage fault detection system

Info

Publication number: CN102607867B
Application number: CN201210042062.XA
Authority: CN
Inventors: 魏秀琨; 贾利民; 柳海; 鞠志强; 张秀洁; 汪煜婷; 郭淑萍; 兰添; 林帅; 郭昆
Original assignee: Beijing Jiaotong University
Current assignee: Beijing Jiaotong University
Priority date: 2012-02-22
Filing date: 2012-02-22
Publication date: 2014-12-10
Anticipated expiration: 2032-02-22
Also published as: CN102607867A

Abstract

The invention relates to an on passage fault detection method based on a GLRT (generalized likelihood ratio test) train suspension system, which comprises the steps of 1) modeling a vehicle suspension system of a train and designing a corresponding Kalman filter; 2) obtaining the acceleration signal at positions during operation of the train by an acceleration sensor; 3) performing anti-aliasing filter, high pass filter and secondary integral pretreatment on the acceleration signal to obtain system output; 4) processing the system output by the Kalman filter to obtain residual output; and 5) judging whether the train breaks down or not by the GLRT method according to the residual output. The on-passage fault detection technology of the suspension system can detect faults of the suspension system with higher robustness and effectively reduce the mistaken reporting and missed reporting rate of fault.

Description

In-transit fault detection system and detection method based on GLRT train suspension system

技术领域technical field

本发明涉及列车故障检测系统，特别是涉及基于GLRT列车悬挂系在途故障检测系统及检测方法。The invention relates to a train fault detection system, in particular to a GLRT-based train suspension system in-transit fault detection system and a detection method.

背景技术Background technique

列车悬挂系统位于列车车体与转向架之间以及转向架与轮对之间，由大量不同的部件构成，包括螺旋弹簧、阻尼器、空气弹簧等。轨道车辆的悬挂系统通常分为一系(位于轮对与转向架之间)与二系(位于转向架与车体之间)，同时也根据其对列车运动状态的影响分为横向与垂向系统。一方面，悬挂系统支撑着车体与转向架；另一方面，悬挂系还起到缓冲由轨道不平顺所引起的轮轨作用力、控制列车行驶方向、保持运行舒适性等作用。The train suspension system is located between the train body and the bogie and between the bogie and the wheelset, and consists of a large number of different components, including coil springs, dampers, air springs, etc. The suspension system of rail vehicles is usually divided into the first series (located between the wheel set and the bogie) and the second series (located between the bogie and the car body). system. On the one hand, the suspension system supports the car body and the bogie; on the other hand, the suspension system also plays the role of buffering the wheel-rail force caused by track irregularities, controlling the direction of the train, and maintaining running comfort.

目前，当前国内外较为成熟的列车在途故障检测系统大多是针对列车动力系、辅助系、制动系等子系统，即牵引电机、逆变器、空调系统、车门系统、空气制动系统的故障检测。即使在针对列车走行系状态监控的技术系统中，也以车轮、轴承、转向架构架等部位为主，几乎没有明确以列车悬挂系统作为状态监控与故障检测对象的系统与方法。At present, most of the more mature train in-transit fault detection systems at home and abroad are aimed at the faults of train power system, auxiliary system, brake system and other subsystems, that is, the faults of traction motor, inverter, air conditioning system, door system, and air brake system. detection. Even in the technical system for the state monitoring of the train running system, the wheels, bearings, bogie frames and other parts are the main parts, and there are almost no systems and methods that clearly use the train suspension system as the object of state monitoring and fault detection.

另一方面，在检测对象包括车辆走行部(列车走行部中包括一系二系悬挂系统)的一些列车故障检测系统中，其检测内容只是简单的表现为转向架的运动状态是否平稳，以一些简单的震动加速度指标或者震动频域指标值作为说明，没有明确悬挂系统是否发生了故障。同时列车走行部的运行状态除了直接受列车悬挂系统性能影响之外，线路条件、车辆负载情况、列车运行工况等都对其产生十分重要的作用，因此以往的检测系统并不能获取导致走行部暂时运动状态失稳的原因。On the other hand, in some train fault detection systems where the detection object includes the running part of the vehicle (the running part of the train includes the primary and secondary suspension systems), the detection content is simply whether the motion state of the bogie is stable or not. The simple vibration acceleration index or the vibration frequency domain index value is used as an illustration, and it is not clear whether the suspension system has failed. At the same time, the running state of the running part of the train is not only directly affected by the performance of the train suspension system, but also the line conditions, vehicle load conditions, and train operating conditions all play a very important role in it. Causes of temporary movement instability.

发明内容Contents of the invention

为避免以上现有技术的不足，本发明提出一种基于GLRT列车悬挂系在途故障检测系统及检测方法，以解决能对列车悬挂系统状态进行实时检测。In order to avoid the above deficiencies in the prior art, the present invention proposes a GLRT-based train suspension system in-transit fault detection system and detection method to solve the problem of real-time detection of the state of the train suspension system.

本发明的目的通过以下技术方案来实现：The purpose of the present invention is achieved through the following technical solutions:

基于GLRT列车悬挂系在途故障检测方法，该方法包括：Based on the in-transit fault detection method of the GLRT train suspension system, the method includes:

1)对列车车辆悬挂系统进行建模并设计相应的Kalman滤波器；1) Model the suspension system of the train vehicle and design the corresponding Kalman filter;

2)利用加速度传感器获取列车运行时各个位置的加速度信号；2) Utilize the acceleration sensor to obtain the acceleration signals of each position when the train is running;

3)对加速度信号进行抗混叠滤波、高通滤波、二次积分预处理，获得系统输出；3) Perform anti-aliasing filtering, high-pass filtering, and secondary integral preprocessing on the acceleration signal to obtain system output;

4)利用Kalman滤波器对所述的系统输出进行处理，获得残差输出；4) using the Kalman filter to process the system output to obtain a residual output;

5)根据残差序列运用GLRT方法判断列车是否发生故障。5) According to the residual sequence, the GLRT method is used to judge whether the train is faulty.

进一步，步骤1)对列车车辆悬挂系统进行建模设计相应的Kalman滤波器具体为：Further, step 1) carries out modeling design corresponding Kalman filter specifically to train vehicle suspension system:

101)根据列车悬挂系统在运行过程中的动力学特性建立悬挂系统横向、垂向动力学模型，构建横向、垂向悬挂系统空间状态方程；101) Establish the lateral and vertical dynamic models of the suspension system according to the dynamic characteristics of the train suspension system during operation, and construct the space state equations of the lateral and vertical suspension systems;

所述系统空间状态方程为： $\begin{matrix} \overset{\cdot}{x} = Ax + B_{d} d \\ y = Cx + D_{d} d \end{matrix}$ The space state equation of the system is: $\begin{matrix} \overset{\cdot}{x} = Ax + B_{d} d \\ the y = Cx + {D.}_{d} d \end{matrix}$

其中x为车辆运动状态变量，d为轨道激励，y为加速度传感器输出信号两次积分后得到的信号，即位移，A,B,C_d,D_d为空间状态方程相应的系数矩阵；Where x is the vehicle motion state variable, d is the track excitation, y is the signal obtained by integrating the output signal of the acceleration sensor twice, that is, the displacement, and A, B, C _d , D _d are the corresponding coefficient matrices of the space state equation;

102)根据系统空间状态方程以及轨道激励信息，以状态估计的误差方差阵的最小为目标设计Kalman滤波器，并得到该滤波器预测误差方差阵P和滤波增益矩阵K,其中102) According to the system space state equation and orbit excitation information, design a Kalman filter with the minimum error variance matrix of state estimation as the goal, and obtain the filter prediction error variance matrix P and filter gain matrix K, where

P＝(A-KC)P(A-KC)′+(B-KD)(B-KD)′P=(A-KC)P(A-KC)'+(B-KD)(B-KD)'

K＝(APC′+BD′)(CPC′+DD′)^-1；K=(APC'+BD')(CPC'+DD') ^-1 ;

公式中A，B，C，D分别是空间状态方程中的A，B_d，C，D_d，P为状态估计的误差方差阵，K为反馈增益矩阵。In the formula, A, B, C, D are A, B _d , C, D _d in the space state equation respectively, P is the error variance matrix of state estimation, and K is the feedback gain matrix.

进一步，步骤5)根据所述滤波器输出残差运用GLRT算法判断列车悬挂系是否发生故障，具体为：当l_k超出一个阈值的时候，即判断系统发生了故障；Further, step 5) using GLRT algorithm to judge whether the train suspension system breaks down according to the residual error of the filter output, specifically: when _lk exceeds a threshold, it is judged that the system has broken down;

${l l}_{k k} = = \underset{{τ τ}^{* *} &Element; &Element; [[k k - - L L,, k k]]}{max max} {λ λ}_{k k} (({τ τ}^{* *})) = = \underset{{τ τ}^{* *} &Element; &Element; [[k k - - L L,, k k]]}{max max} \frac{{x x}_{k k}^{22} (({τ τ}^{* *}))}{{S S}_{k k} (({τ τ}^{* *}))} \approx \approx \frac{{x x}_{k k}^{22}}{{S S}_{k k}}$

${x x}_{k k} (({τ τ}^{* *})) = = {Σ Σ}_{j j = = {τ τ}^{* *}}^{k k} {G G}_{j j}^{' '} (({τ τ}^{* *})) {R R}_{j j}^{- - 11} {ρ ρ}_{j j}$

R_j＝C_jPC′_j+D_jD′_j R _j ＝C _j PC′ _j +D _j D′ _j

${S S}_{k k} (({τ τ}^{* *})) = = {Σ Σ}_{j j = = {τ τ}^{* *}}^{k k} {G G}_{j j}^{' '} (({τ τ}^{* *})) {R R}_{j j}^{- - 11} {G G}_{j j}$

其中P为Kalman滤波器误差方差阵；ρ_j为Kalman滤波器输出残差；G_j为故障特征向量。Among them, P is the error variance matrix of Kalman filter; ρ _j is the output residual of Kalman filter; G _j is the fault feature vector.

进一步，该系统包括：Further, the system includes:

传感器，用于获取列车在各个位置的加速度信息；The sensor is used to obtain the acceleration information of the train at various positions;

数据采集单元，负责连接传感器和数据预处理单元，将传感器发送的模拟信号转换为数据预处理单元可以识别的格式，以统一的通信协议传输信号给数据预处理单元，实现各传感器测量数据的采集和转换处理；The data acquisition unit is responsible for connecting the sensor and the data preprocessing unit, converting the analog signal sent by the sensor into a format that the data preprocessing unit can recognize, and transmitting the signal to the data preprocessing unit with a unified communication protocol to realize the collection of measurement data of each sensor and conversion processing;

数据预处理单元，与所述信号采集单元相连，对信号采集单元的数据进行管理，且接收数据采集单元送来的数据及对数据进行预处理，包括高通滤波、二次积分处理工作，然后再将预处理结果通过以太网传给故障诊断单元；The data preprocessing unit is connected with the signal acquisition unit, manages the data of the signal acquisition unit, and receives the data sent by the data acquisition unit and preprocesses the data, including high-pass filtering and secondary integral processing, and then Send the preprocessing results to the fault diagnosis unit via Ethernet;

故障诊断单元，对接收到的数据预处理结果进行分析，判断列车悬挂系统是否发生故障。The fault diagnosis unit analyzes the received data preprocessing results to determine whether the train suspension system is faulty.

进一步，所述的传感器包括车体传感器和转向架传感器，分别用于获取车体XYZ三项加速度信号与转向架YZ二项加速度信号。Further, the sensors include a car body sensor and a bogie sensor, which are respectively used to obtain the car body XYZ three-term acceleration signal and the bogie YZ two-term acceleration signal.

进一步，所述的车体传感器位于列车底板四角。Further, the vehicle body sensors are located at four corners of the train floor.

进一步，所述的转向架传感器位于列车的转向架侧梁上方两端。Further, the bogie sensors are located at both ends above the bogie side beams of the train.

进一步，所述车辆运行状态包括车体与转向架的沉浮运动状态、点头运动状态、侧滚运动状态。Further, the running state of the vehicle includes the ups and downs of the vehicle body and the bogie, the nodding motion state, and the roll motion state.

本发明的优点在于：The advantages of the present invention are:

本发明提出的列车悬挂系在途故障检测系统直接以列车悬挂系统作为检测对象，该检测技术对线路条件、列车运行工况等环境干扰具有鲁棒性，而对悬挂系自身故障却十分敏感，在悬挂系统发生较小故障的时候就可以实时地检测到故障的存在并报警。The in-transit fault detection system of the train suspension system proposed by the present invention directly uses the train suspension system as the detection object. This detection technology is robust to environmental interference such as line conditions and train operating conditions, but is very sensitive to the failure of the suspension system itself. When a minor fault occurs in the suspension system, the existence of the fault can be detected in real time and an alarm is given.

附图说明Description of drawings

图1：本发明故障检测系统结构框架；Fig. 1: structural framework of fault detection system of the present invention;

图2：传感器布设位置；Figure 2: Sensor layout location;

图3：数据采集单元信号调理电路流程；Figure 3: Process flow of the signal conditioning circuit of the data acquisition unit;

图4：数据采集单元电路框图；Figure 4: Block diagram of the data acquisition unit circuit;

图5：数据预处理单元数据处理逻辑如；Figure 5: The data processing logic of the data preprocessing unit such as;

图6：车辆垂向悬挂系统示意图；Figure 6: Schematic diagram of the vertical suspension system of the vehicle;

图7：车辆横向悬挂系统示意图Figure 7: Schematic diagram of the vehicle lateral suspension system

图8为卡尔曼滤波器滤波估计流程图；Fig. 8 is a flow chart of Kalman filter filtering estimation;

图9为卡尔曼滤波增益和误差方差阵计算流程图；Fig. 9 is the calculation flowchart of Kalman filter gain and error variance matrix;

图10：Kalman滤波器滤波产生残差；Figure 10: Kalman filter filtering produces residuals;

图11：空璜故障检测算法验证图示。Figure 11: Schematic illustration of the validation of the Air Juan Fault Detection Algorithm.

具体实施方式Detailed ways

如图1所示为本发明故障检测系统结构框架图，所述故障检测系统包括：传感器，用于获取列车在各个位置的加速度信息；数据采集单元，负责连接传感器和数据预处理单元，将传感器发送的模拟信号转换为数据预处理单元可以识别的格式，以统一的通信协议发送到数据预处理单元，实现各传感器测量数据的采集和转换理；数据预处理单元，逻辑上负责各数据采集单元的数据管理及车辆网络管理，且接收数据采集单元送来的数据，对数据进行坐标变换、高通滤波、二次积分运算等工作，然后再将预处理结果通过以太网传给故障诊断单元；所述数据采集单元与预处理单元在每个车厢上都有安装，而故障诊断主机一辆列车配备一台，即故障诊断单元，通过以太网搜集各个车厢上采集的信息，对接收到的数据预处理结果进行判断，判断列车悬挂系统是否发生故障。As shown in Figure 1, it is a structural frame diagram of the fault detection system of the present invention, and the fault detection system includes: a sensor for obtaining the acceleration information of the train at various positions; a data acquisition unit is responsible for connecting the sensor and the data preprocessing unit, and the sensor The sent analog signal is converted into a format recognizable by the data preprocessing unit, and sent to the data preprocessing unit with a unified communication protocol to realize the collection and conversion of the measurement data of each sensor; the data preprocessing unit is logically responsible for each data acquisition unit Data management and vehicle network management, and receive the data sent by the data acquisition unit, perform coordinate transformation, high-pass filtering, and secondary integral calculation on the data, and then transmit the preprocessing results to the fault diagnosis unit through Ethernet; The above-mentioned data collection unit and preprocessing unit are installed on each car, and a fault diagnosis host is equipped with one train, that is, the fault diagnosis unit, which collects the information collected on each car through Ethernet, and predicts the received data. The processing result is judged to judge whether the train suspension system fails.

下面对系统各单元分别进行详细说明。Each unit of the system will be described in detail below.

如图2所示，所述传感器类型有两种，为车体传感器与转向架传感器，分别用于获取车体XYZ三项加速度信号与转向架YZ二项加速度信号。车体传感器布设在车底板的四个角，与车底板边缘距离均为400mm。该布设方案不仅能有效获得车体各个位置各方向加速度信息，且能够用于转换摇头、点头、侧滚等角加速度，全面获取车体各运动状态数据，为故障检测、分离提供充分数据。转向架传感器布设在转向架构架上方四角位置，获取转向架横向以及垂向加速度信号。这样的布设方案同样能够最大限度的获取转向架的运动状态数据。As shown in Fig. 2, there are two types of sensors, the car body sensor and the bogie sensor, which are respectively used to obtain the XYZ three-term acceleration signal of the car body and the YZ two-term acceleration signal of the bogie. The car body sensors are arranged at the four corners of the car floor, and the distance from the edge of the car floor is 400mm. This layout scheme can not only effectively obtain the acceleration information of each position and direction of the car body, but also can be used to convert angular acceleration such as shaking head, nodding, and roll, and comprehensively obtain the data of each motion state of the car body, providing sufficient data for fault detection and separation. The bogie sensors are arranged at the four corners above the bogie frame to obtain the lateral and vertical acceleration signals of the bogie. Such a layout scheme can also obtain the motion state data of the bogie to the maximum extent.

数据采集单元负责连接传感器和数据预处理单元，将传感器发送的模拟信号转换为数据预处理单元可以识别的格式，以统一的通信协议向上发送，实现各传感器测量数据的采集和转换理。数据采集单元具体将完成以下工作：电流信号到电压信号的转换，抗混叠滤波、电压转换，模拟信号的A/D转换等工作，并通过以太网将采集数据传给数据预处理单元。The data acquisition unit is responsible for connecting the sensor and the data preprocessing unit, converting the analog signal sent by the sensor into a format that the data preprocessing unit can recognize, and sending it upwards with a unified communication protocol to realize the collection and conversion of the measurement data of each sensor. The data acquisition unit will specifically complete the following work: conversion of current signal to voltage signal, anti-aliasing filter, voltage conversion, A/D conversion of analog signal, etc., and transmit the collected data to the data preprocessing unit through Ethernet.

数据采集单元电路流程及框图如图3和4所示，数据采集单元具体规格如下：The circuit flow and block diagram of the data acquisition unit are shown in Figures 3 and 4. The specific specifications of the data acquisition unit are as follows:

√采用16位AD转换电路，每个通道能达到200KSPS采样速率，采用二阶抗混叠低通滤波器，提供数字滤波器实现过采样功能√ Adopt 16-bit AD conversion circuit, each channel can reach 200KSPS sampling rate, adopt second-order anti-aliasing low-pass filter, provide digital filter to realize over-sampling function

√采用10M/100M/1000M以太网口传输数据，支持IEE1588网络同步协议√ Adopt 10M/100M/1000M Ethernet port to transmit data, support IEE1588 network synchronization protocol

√CPU采用Freescale MPC系列高性能处理器，主频400M，内存不小于128MB，板载64M flash√The CPU adopts Freescale MPC series high-performance processors, the main frequency is 400M, the memory is not less than 128MB, and the onboard 64M flash

√Altera Cyclone EPC4系列FPGA作为协处理器，可用于采集数据预处理运算√Altera Cyclone EPC4 series FPGA as a coprocessor can be used to collect data preprocessing operations

√24VDC供电，单板功耗≤15W√24VDC power supply, board power consumption ≤15W

√全部采用高可靠性工业级宽温器件√ All adopt high-reliability industrial-grade wide-temperature devices

所述信号预处理单元，物理上负责连接信号采集单元与网络节点，逻辑上负责各信号采集单元的数据管理及车辆网络管理。信号预处理单元接收汇聚信号采集单元送来的数据，对数据进行坐标变换、高通滤波、二次积分运算等工作，然后再将预处理结果通过以太网传给诊断服务主机。The signal preprocessing unit is physically responsible for connecting the signal acquisition unit and the network node, and logically responsible for data management of each signal acquisition unit and vehicle network management. The signal preprocessing unit receives the data sent by the aggregation signal acquisition unit, performs coordinate transformation, high-pass filtering, and quadratic integral operation on the data, and then transmits the preprocessing results to the diagnostic service host through Ethernet.

数据处理板的主要功能是对数据进行数字化滤波，积分运算、特征提取等工作，以及完成来自不同传感器接入的异构数据的整合功能。The main function of the data processing board is to digitally filter the data, integrate calculations, feature extraction, etc., and complete the integration of heterogeneous data from different sensors.

数据预处理单元数据处理逻辑如图5所示：The data processing logic of the data preprocessing unit is shown in Figure 5:

●数据处理板规格●Data processing board specification

√CPU采用Freescale MPC系列高性能处理器，主频400M√CPU adopts Freescale MPC series high-performance processor, the main frequency is 400M

√内存不小于128MB√ Memory not less than 128MB

√板载64M flash√Onboard 64M flash

√采用10M/100M/1000M以太网口传输数据√ Use 10M/100M/1000M Ethernet port to transmit data

√5VDC供电，单板功耗≤15W√5VDC power supply, board power consumption ≤15W

故障诊断单元，由车载高性能工业计算机构成，对接收到的数据预处理结果进行判断，判断列车悬挂系统是否发生故障。具体诊断方法如下。The fault diagnosis unit is composed of a vehicle-mounted high-performance industrial computer, which judges the preprocessing results of the received data and judges whether the train suspension system is faulty. The specific diagnosis method is as follows.

1.对列车悬挂系统进行建模并设计相应的Kalman滤波器。1. Model the train suspension system and design the corresponding Kalman filter.

如图6,7所示，对车辆垂向、横向悬挂系统建模，根据车辆运动的动力学方程分别建立车体悬挂系统垂向、横向微分方程模型，进而导出车辆悬挂系统的空间状态方程，其形式为： $\begin{matrix} \overset{\cdot}{x} = Ax + B_{d} d \\ y = Cx + D_{d} d \end{matrix}$ As shown in Figures 6 and 7, the vertical and lateral suspension systems of the vehicle are modeled, and the vertical and lateral differential equation models of the vehicle body suspension system are respectively established according to the dynamic equations of the vehicle motion, and then the space state equation of the vehicle suspension system is derived. Its form is: $\begin{matrix} \overset{\cdot}{x} = Ax + B_{d} d \\ the y = Cx + {D.}_{d} d \end{matrix}$

其中x为状态变量，d为轨道激励，y为加速度传感器输出后两次积分量，即位移,A,B,C_d,D_d为空间状态方程相应的系数矩阵。Among them, x is the state variable, d is the track excitation, y is the output of the accelerometer after two integrals, that is, the displacement, and A, B, C _d , D _d are the corresponding coefficient matrices of the space state equation.

如图8为卡尔曼滤波器滤波估计流程图，图9为卡尔曼滤波增益和误差方差阵计算流程图。卡尔曼滤波器进行波形估计的原则为实现状态估计的误差方差阵的最小化，得到滤波器预测误差方差阵P为：FIG. 8 is a flow chart of Kalman filter estimation, and FIG. 9 is a flow chart of Kalman filter gain and error variance matrix calculation. The principle of Kalman filter for waveform estimation is to minimize the error variance matrix of state estimation, and the filter prediction error variance matrix P is obtained as:

P＝(A-KC)P(A-KC)′+(B-KD)(B-KD)′P=(A-KC)P(A-KC)'+(B-KD)(B-KD)'

滤波增益矩阵K的计算公式为：The calculation formula of the filter gain matrix K is:

K＝(APC′+BD′)(CPC′+DD′)^-1 K=(APC'+BD')(CPC'+DD') ^-1

2.利用加速度传感器获取列车运行时各个位置的加速度信号。2. Use the acceleration sensor to obtain the acceleration signals of each position when the train is running.

3.对加速度信号进行抗混叠滤波、高通滤波、二次积分预处理，获得系统输出。3. Perform anti-aliasing filtering, high-pass filtering, and quadratic integral preprocessing on the acceleration signal to obtain system output.

4.利用Kalman滤波器对所述的系统输出进行处理，获得残差输出，如图10中虚线框内部分。4. Use the Kalman filter to process the system output to obtain the residual output, as shown in the part inside the dotted line box in FIG. 10 .

5.根据残差序列运用GLRT方法判断列车是否发生故障。5. Use the GLRT method to judge whether the train is faulty or not according to the residual sequence.

此处是基于GLRT的故障检测，GLRT方法即广义概率比率检验方法(Generalized Likelihood Ratio Test)，其基本原理为在每一个时间窗内，利用卡尔曼滤波器输出的残差计算当前系统故障、无故障的条件概率比值，并以该比值作为系统是否发生故障的指标，当下面公式中l_k超出一个预先设定的阈值时，即判断系统发生了故障。Here is the fault detection based on GLRT. The GLRT method is the generalized probability ratio test method (Generalized Likelihood Ratio Test). Its basic principle is to use the residual output of the Kalman filter to calculate the current system fault, The ratio of the conditional probability of failure is used as an indicator of whether the system fails. When l _k in the following formula exceeds a preset threshold, it is judged that the system has failed.

广义概率比率公式为：The generalized probability ratio formula is:

其中：in:

R_j＝C_jPC′_j+D_jD′_j R _j ＝C _j PC′ _j +D _j D′ _j

上述公式中符号含义：P为Kalman滤波器误差方差阵；ρ_j为Kalman滤波器输出残差；G_j为故障特征向量。Meanings of the symbols in the above formula: P is the Kalman filter error variance matrix; _ρj is the Kalman filter output residual; _Gj is the fault feature vector.

我们利用SIMPACK动力学仿真专用软件对上述故障检测方法进行验证，车辆二系空璜正常刚度值为0.26513MN/m，现在系统运行至40s时发生故障，漏气导致空璜刚度值变为正常值的一半，即0.13256MN/m。如图11可以看到我们的检测算法迅速地检测到了这一故障的发生。当然，发生故障的悬挂系部件也可以是一系二系的其他弹簧与阻尼。经大量的仿真试验验证了该故障检测算法的可靠性。本发明这种悬挂系在线故障检测技术可以以较高的鲁棒性进行悬挂系的故障检测，有效降低故障误报漏报率。本发明列车悬挂系在途故障检测系统是针对列车悬挂系统进行状态监控、故障检测的全新的方法，与已有的列车故障检测系统或者列车状态监控系统有本质的不同。We used SIMPACK dynamic simulation software to verify the above fault detection method. The normal stiffness value of the second-series air jumbo of the vehicle is 0.26513MN/m. Now the system fails when it runs to 40s, and the air leakage causes the air jumbo stiffness value to become a normal value. Half of that, namely 0.13256MN/m. As shown in Figure 11, we can see that our detection algorithm quickly detected the occurrence of this fault. Of course, the suspension components that fail can also be other springs and dampers of the first series and the second series. The reliability of the fault detection algorithm is verified by a large number of simulation experiments. The on-line fault detection technology of the suspension system of the present invention can detect the fault of the suspension system with high robustness, effectively reducing the rate of fault false positives and negative negatives. The in-transit fault detection system of the train suspension system of the present invention is a brand-new method for state monitoring and fault detection of the train suspension system, which is essentially different from the existing train fault detection system or train state monitoring system.

应当理解，以上借助优选实施例对本发明的技术方案进行的详细说明是示意性的而非限制性的。本领域的普通技术人员在阅读本发明说明书的基础上可以对各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。It should be understood that the above detailed description of the technical solution of the present invention with the aid of preferred embodiments is illustrative rather than restrictive. Those skilled in the art can modify the technical solutions recorded in each embodiment on the basis of reading the description of the present invention, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the corresponding technical solutions Essentially deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. based on GLRT train suspension system fault detection method in transit, it is characterized in that, the method comprises:

1) Model the suspension system of the train vehicle and design the corresponding Kalman filter;

2) Utilize the acceleration sensor to obtain the acceleration signals of each position when the train is running;

3) Perform anti-aliasing filtering, high-pass filtering, and secondary integral preprocessing on the acceleration signal to obtain system output;

4) using the Kalman filter to process the system output to obtain a residual output;

5) Use the GLRT method to judge whether the train breaks down according to the residual sequence;

Described step 1) carries out modeling design corresponding Kalman filter specifically to train vehicle suspension system:

101) Establish the lateral and vertical dynamic models of the suspension system according to the dynamic characteristics of the train suspension system during operation, and construct the space state equation of the lateral and vertical suspension systems;

The space state equation of the system is:

\begin{matrix} \overset{&Center Dot;}{x} = Ax + B_{d} d \\ the y = Cx + {D.}_{d} d \end{matrix}

Among them, x is the vehicle motion state variable, d is the track excitation, y is the signal obtained after the acceleration sensor output signal is integrated twice, that is, the displacement, A, B _d , C, D _d are the corresponding coefficient matrices of the space state equation;

102) According to the system space state equation and orbital excitation information, design a Kalman filter with the minimum error variance matrix of state estimation as the goal, and obtain the filter prediction error variance matrix P and filter gain matrix K, where

P=(A-KC)P(A-KC)'+(B-KD)(B-KD)'

K=(APC'+BD')(CPC'+DD') ^-1 ;

In the formula, A, B, C, D are A, B _d , C, D _d in the space state equation respectively, P is the error variance matrix of state estimation, and K is the feedback gain matrix;

Described step 5) use GLRT algorithm to judge whether train suspension system breaks down according to described filter output residual error, specifically: when _lk exceeds a threshold value, namely judgment system broke down;

{l l}_{k k} = = \underset{{τ τ}^{* *} &Element; &Element; [[k k - - L L,, k k]]}{max max} {λ λ}_{k k} (({τ τ}^{* *})) = = \underset{{τ τ}^{* *} &Element; &Element; [[k k - - L L,, k k]]}{max max} \frac{{x x}_{k k}^{22} (({τ τ}^{* *}))}{{S S}_{k k} (({τ τ}^{* *}))} \approx \approx \frac{{x x}_{k k}^{22}}{{S S}_{k k}}

{x x}_{k k} (({τ τ}^{* *})) = = {Σ Σ}_{j j = = {τ τ}^{* *}}^{k k} {G G}_{j j}^{' '} (({τ τ}^{* *})) {R R}_{j j}^{- - 11} {ρ ρ}_{j j}

R _j ＝C _j PC′ _j +D _j D′ _j

{S S}_{k k} (({τ τ}^{* *})) = = {Σ Σ}_{j j = = {τ τ}^{* *}}^{k k} {G G}_{j j}^{' '} (({τ τ}^{* *})) {R R}_{j j}^{- - 11} {G G}_{j j}

Among them, P is the error variance matrix of Kalman filter; ρ _j is the output residual of Kalman filter; G _j is the fault feature vector.

2. The in-transit fault detection system of the train suspension system based on GLRT is characterized in that the system includes:

The sensor is used to obtain the acceleration information of the train at various positions;

The data acquisition unit is responsible for connecting the sensor and the data preprocessing unit, converting the analog signal sent by the sensor into a format that the data preprocessing unit can recognize, and transmitting the signal to the data preprocessing unit with a unified communication protocol to realize the collection of measurement data of each sensor and conversion processing;

The data preprocessing unit is connected with the signal acquisition unit, manages the data of the signal acquisition unit, and receives the data sent by the data acquisition unit and preprocesses the data, including high-pass filtering and secondary integral processing, and then Send the preprocessing results to the fault diagnosis unit via Ethernet;

Fault diagnosis unit, including:

According to the dynamic characteristics of the train suspension system during operation, the lateral and vertical dynamic models of the suspension system are established, and the space state equations of the lateral and vertical suspension systems are constructed;

The space state equation of the system is:

\begin{matrix} \overset{\cdot}{x} = Ax + B_{d} d \\ the y = Cx + {D.}_{d} d \end{matrix}

Where x is the vehicle motion state variable, d is the track excitation, y is the signal obtained by integrating the output signal of the acceleration sensor twice, that is, the displacement, and A, B _d , C, D _d are the corresponding coefficient matrices of the space state equation;

According to the system space state equation and orbit excitation information, the Kalman filter is designed with the goal of minimizing the error variance matrix of state estimation, and the filter prediction error variance matrix P and filter gain matrix K are obtained, where

P=(A-KC)P(A-KC)'+(B-KD)(B-KD)'

K=(APC'+BD')(CPC'+DD') ^-1 ;

Use the Kalman filter to process the preprocessed data to obtain the residual, and use the GLRT algorithm to judge whether the train suspension system is faulty, specifically: when l _k exceeds a threshold, it is judged that the system is faulty;

{l l}_{k k} = = \underset{{τ τ}^{* *} &Element; &Element; [[k k - - L L,, k k]]}{max max} {λ λ}_{k k} (({τ τ}^{* *})) = = \underset{{τ τ}^{* *} &Element; &Element; [[k k - - L L,, k k]]}{max max} \frac{{x x}_{k k}^{22} (({τ τ}^{* *}))}{{S S}_{k k} (({τ τ}^{- - * *}))} \approx \approx \frac{{x x}_{k k}^{22}}{{S S}_{k k}}

{x x}_{k k} (({τ τ}^{* *})) = = {Σ Σ}_{j j = = {τ τ}^{* *}}^{k k} {G G}_{j j}^{' '} (({τ τ}^{* *})) {R R}_{j j}^{- - 11} {ρ ρ}_{j j}

R _j ＝C _j PC′ _j +D _j D′ _j

{S S}_{k k} (({τ τ}^{* *})) = = {Σ Σ}_{j j = = {τ τ}^{* *}}^{k k} {G G}_{j j}^{' '} (({τ τ}^{* *})) {R R}_{j j}^{- - 11} {G G}_{j j}

3. The in-transit fault detection system for train suspension systems based on GLRT according to claim 2, wherein said sensors include car body sensors and bogie sensors, which are respectively used to obtain car body XYZ three acceleration signals and steering Frame YZ binomial acceleration signal.

4. The GLRT-based train suspension system in-transit fault detection system according to claim 3, characterized in that, the car body sensors are located at the four corners of the train floor.

5. The in-transit fault detection system for train suspension system based on GLRT according to claim 3, characterized in that, the bogie sensors are located at both ends above the bogie side beams of the train.

6. The in-transit fault detection system for train suspension system based on GLRT according to claim 2, characterized in that the vehicle running state includes the ups and downs of the car body and the bogie, the nodding motion state, and the roll motion state.